Chapter 1: Compressed Like a DJ
Before I got into software, I used to make beats in high school. A lot of what I learned staying up all night creating music led to my love for software development. Regardless, it’s undeniable what the rise of electronic music did for the industry.
Way back when, music was played by musicians on physical instruments. They dedicated years of their lives to their craft, chasing perfection. Then it all got computerized. Orchestras were reduced to producers sitting in front of million-dollar technical behemoths, able to pull any sound out of thin air, manipulate it, and compose masterpieces in their own right. Musical labor got consolidated into machines, and talent got compressed. One person could piece together a composition that would have previously taken an entire orchestra, hours of organization, and hundreds of years of combined experience across some of the most talented musicians in the world [23]. The output of a singular savant could increase a thousandfold due to the consolidation and computerization of labor.
| Music (Pre-Digital) | Music (Post-Digital) | Software (Pre-AI) | Software (Post-AI) | |
|---|---|---|---|---|
| Who builds | Orchestra (50+ musicians) | Solo producer with a DAW | Dev team (5-15 engineers) | FDE with AI tools |
| Skill distribution | Spread across specialists | Compressed into one person | Spread across roles | Compressed into one person |
| Output speed | Weeks of rehearsal + recording | Hours to days | Months of sprints | Days to weeks |
| Bottleneck | Coordination + talent | Creative vision + taste | Implementation | Planning + the what |
The Compression Effect
BEFORE AFTER ┌─────────────────────────┐ ┌─────────────────────────┐ │ │ │ │ │ 50+ Musicians │ │ Solo Producer + DAW │ │ Conductor │ ─────► │ │ │ Studio Team │ │ 1 person, │ │ Sound Engineer │ │ same output │ │ Years of rehearsal │ │ │ │ │ │ │ └─────────────────────────┘ └─────────────────────────┘ ┌─────────────────────────┐ ┌─────────────────────────┐ │ │ │ │ │ 5-15 Engineers │ │ FDE + AI Tools │ │ PM / Scrum Master │ ─────► │ │ │ Designer │ │ 1 person, │ │ QA Team │ │ orchestrating agents │ │ Months of sprints │ │ │ │ │ │ │ └─────────────────────────┘ └─────────────────────────┘
There’s a reason the term “orchestrating AI” has become so prevalent. The same transformative process is happening right now in the software industry. Just like musicians, you have this cohort of craftsmen dedicating years of their lives to the art of software. Beautiful, elegant code backed by decades of experience and thousands of hours poring over documentation is now getting compressed into generative AI outputs. And just like with music, do you lose that uncanny sense of craftsmanship? Of course you do. There will still be a place for it, just like live orchestras still exist today. But by and large, a shift toward a singular conductor — able to pull in the efforts of a generation of craftsmen — is transforming the landscape. The controller, the producer or DJ’s experience, is now center stage.
As I write this, I have no idea if that role itself will be replaced. Perhaps it will be in three months at the rate we’re moving. But that’s the current landscape. From our perspective, just like we’ve seen before, talent has been compressed. So what does that mean, and what are the outcomes of this transformation?
It shifts the software development landscape to one that is planning-heavy. I saw a tweet that summed it up perfectly: AI has sped up coding time, not thinking time. This observation is backed by growing research — a 2025 METR study found that experienced open-source developers actually took 19% longer to complete tasks when using AI tools, despite believing AI had sped them up by 20% [21]. Addy Osmani, an engineering manager at Google, formalized this as “The 70% Problem” — AI can rapidly produce 70% of a solution, but the remaining 30%, the hard thinking, remains as challenging as ever [22]. Never has a ruthless and transparent planning and discovery process been more important. Everything on the implementation side is running on an exponential curve now. If you give AI bad plans, it’s going to hurdle you in the wrong direction ten times faster than before. But the opposite is true as well. Code is getting written faster than ever.
This echoes what the research has been telling us for decades. The Standish Group’s CHAOS reports, spanning over 50,000 projects globally, have consistently found that roughly 66% of technology projects end in partial or total failure [1]. McKinsey’s research, conducted with the University of Oxford across 5,400 large-scale IT projects, found that these projects run an average of 45% over budget and deliver 56% less value than predicted — and that 17% go so badly they threaten the very existence of the company [2]. The root causes aren’t new either: unclear requirements and poor planning are cited as leading contributors in 39% to 80% of project failures, depending on the study [3][4]. The difference now is that AI has dramatically accelerated the consequences of getting it wrong.
We have a situation in which we’ve moved from horses to race cars. Yes, you can get from point A to point B faster, but crashes are more lethal.
Speed vs. Risk: The New Tradeoff
Speed of Development ───────────────────────────────────────────► ┌──────────────────────────────────────────────┐ HORSE & BUGGY │ ▓▓▓▓░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ │ (Pre-AI) │ │ │ Risk: Low Recovery: Slow │ ├──────────────────────────────────────────────┤ RACE CAR │ ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ │ (Post-AI) │ │ │ Risk: HIGH Recovery: FAST │ └──────────────────────────────────────────────┘ Key Insight: You CAN recover faster — but you must avoid catastrophic crashes through proper planning.
That said, we’ve also observed that if you take a wrong turn, you can get back on track faster than before. Going back to the musical analogy: if you have a verse that is out of key, you don’t need to reproduce the entire song. You can just re-record that specific section. It would be disingenuous for me to say that if you don’t get the planning phase completely right, you have to restart. I’m claiming the opposite. We view a project as a race car — not a bullet train where a slight hiccup can derail the entire trip. What we want to avoid is crashes. Before we rev up this F1 car: are there trees in the way? Can we get to the finish line? Sure, we can’t see what’s around the second curve, but we understand that we’re not going to die. We can get there, and get there fast. And if we go off course, we slow down, realign, and hit the gas again.
It’s also worth acknowledging that in software development, when you crash, you don’t necessarily die. You could argue that the faster you crash, the faster you figure out feasibility. I think that works on personal projects — projects without a budget, fun projects. But when we’re talking real stakes, cost implications, and needed business outcomes, it’s better to de-risk. A 2024 study by Junade Ali and J.L. Partners, surveying 600 software engineers, found that projects with clear requirements before development started were 97% more likely to succeed, and projects with a documented specification were 50% more likely to succeed [5]. If you’re writing in pencil, go for it. Writing in pen? Probably best to figure out a couple things before you start.
By the Numbers: Why Planning Matters More Than Ever
| Statistic | Source |
|---|---|
| 66% of technology projects end in partial or total failure | Standish Group CHAOS Report [1] |
| 45% over budget on average for large-scale IT projects | McKinsey / Oxford [2] |
| 17% of IT projects threaten the very existence of the company | McKinsey / Oxford [2] |
| 97% more likely to succeed with clear requirements | Ali & J.L. Partners [5] |
| 19% slower when experienced devs use AI (despite feeling faster) | METR Study [21] |
| 70% of a solution is easy with AI; the last 30% is just as hard | Addy Osmani / Google [22] |
Continuing on this: if we get to a point where implementation times and costs go to near zero — which is a real possibility — you could run parallel agents, almost in a simulation, to discover roadblocks and pick the winner. I think that could happen, and it would be a service I would love to offer one day. But that still doesn’t change the need for the core purpose of planning and discovery: determine what we need to build. If we can simulate and determine the risks of how to build it autonomously one day, tremendous. But that’s downstream from what.
This concept is clearly exemplified in a recent article by MIT that explains why 95% of AI initiatives fail. Below is the executive summary of The Gen AI Divide: State of AI in 2025 [24]:
Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result: 95% of organizations are getting zero return. The outcomes are so starkly divided across both buyers (enterprises, mid-market, SMBs) and builders (startups, vendors, consultancies) that we call it the GenAI Divide. Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact.
This divide does not seem to be driven by model quality or regulation, but by approach. Tools like ChatGPT and Copilot are widely adopted. Over 80 percent of organizations have explored or piloted them, and nearly 40 percent report deployment. But these tools primarily enhance individual productivity, not P&L performance.
Meanwhile, enterprise-grade systems — custom or vendor-sold — are being quietly rejected. Sixty percent of organizations evaluated such tools, but only 20 percent reached pilot stage and just 5 percent reached production. Most fail due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations. From interviews, surveys, and analysis of 300 public implementations, four patterns emerged that define the GenAI Divide:
- Limited disruption: Only 2 of 8 major sectors show meaningful structural change
- Enterprise paradox: Big firms lead in pilot volume but lag in scale-up
- Investment bias: Budgets favor visible, top-line functions over high-ROI back office
- Implementation advantage: External partnerships see twice the success rate of internal builds
The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. Most GenAI systems do not retain feedback, adapt to context, or improve over time. A small group of vendors and buyers are achieving faster progress by addressing these limitations directly. Buyers who succeed demand process-specific customization and evaluate tools based on business outcomes rather than software benchmarks. They expect systems that integrate with existing processes and improve over time. Vendors meeting these expectations are securing multi-million-dollar deployments within months. While most implementations don’t drive headcount reduction, organizations that have crossed the GenAI Divide are beginning to see selective workforce impacts in customer support, software engineering, and administrative functions. In addition, the highest-performing organizations report measurable savings from reduced BPO spending and external agency use, particularly in back-office operations. Others cite improved customer retention and sales conversion through automated outreach and intelligent follow-up systems. These early results suggest that learning-capable systems, when targeted at specific processes, can deliver real value — even without major organizational restructuring.
The GenAI Divide at a Glance
┌──────────────────────────────────────────────────────────────────────────┐ │ $30-40 BILLION invested in GenAI │ │ │ ├───────────────────────────────────┬──────────────────────────────────────┤ │ 95% of orgs │ 5% of orgs │ │ │ │ │ ┌─────────────────────────────┐ │ ┌────────────────────────────────┐ │ │ │ │ │ │ │ │ │ │ Zero measurable │ │ │ Millions in extracted │ │ │ │ P&L impact │ │ │ value │ │ │ │ │ │ │ │ │ │ │ • Individual productivity │ │ │ • Process-specific │ │ │ │ only │ │ │ customization │ │ │ │ │ │ │ │ │ │ │ • Brittle workflows │ │ │ • Business outcome focused │ │ │ │ │ │ │ │ │ │ │ • No contextual learning │ │ │ • Learning-capable systems │ │ │ │ │ │ │ │ │ │ └─────────────────────────────┘ │ │ • External partnerships │ │ │ │ │ │ │ │ │ └────────────────────────────────┘ │ └───────────────────────────────────┴──────────────────────────────────────┘ Core barrier to scaling: NOT infrastructure, regulation, or talent — it's LEARNING and determining WHAT to build.
Incredible insight. I want to key in on a couple points here.
First: “But these tools primarily enhance individual productivity, not P&L performance.” This just makes sense. Think about how much work ChatGPT assists with under your direction — drafting emails, summarizing documents, transcribing calls. But would you trust ChatGPT to independently and proactively find jobs and apply to them on your behalf? Maybe. But would you trust it as much as you would trust an employee? I’m not claiming these tools aren’t capable of such tasks. In fact, with the rise of OpenClaw and other prospective fully autonomous bots, I see it as a true possibility relatively soon. But as of right now, a human in the loop is needed. We see this later in the report as well: much of the real impact was realized in BPO and back offices — the mechanical, laborious parts of the organization — places where labor implementation is either extremely systematized or easily directed by a person.
The core point I’m making is that there is clearly value to be derived from these tools, but the failures came from the what, not the how. A clear understanding of the business problems, a clear understanding of model capabilities, and a clear implementation roadmap to determine feasibility — that’s what separates the 5% from the 95%.
Will these models ever be able to determine the what? Honestly, maybe. But the collapse will move up the organizational hierarchy. As we’re seeing from the article, BPO, back office, and software development are getting impacted first — the implementers. Perhaps it stops there. Perhaps it doesn’t. Middle management could collapse, and you could have agentic-led companies or one-person behemoths — a trend already emerging, as the share of new U.S. startups founded by solo entrepreneurs surged from 22% in 2015 to 38% in 2024 [32]. But transparently, that would be such a global disruption that I believe it’s one of those situations where we’ll have to see what the world looks like at that point. If agents can run efficient billion-dollar organizations, humanity as a whole will be disrupted. Issues like agentic rights, consciousness, and organic versus manipulated agentic motivations will be at the forefront of everyone’s mind. We’ll know when — or if — we get there.
As of right now, like with music, the compression of labor and implementation due to AI has created a landscape in which implementation is no longer the bottleneck. Implementation planning is quickly reducing in importance. Thinking and determining the what is the bottleneck.
Chapter 1 Key Takeaways